Hacker News Reader

You don't benchmark foundation model against RLHF model, results aren't very useful.

This does seem to be a RLHF model, not a base model. Unless 'supervised fine-tuning' and 'human preference' mean something else.

Ah I see there is also a llama-2-chat model.

Check out figures 1 & 2 in the Llama-2 paper :) They benchmark against ChatGPT for helpfulness and harmfulness

See figure-2

Looks like it comes in just under GPT-3.5

That is unrelated. Stop spreading misinformation.

It is for the old version and not this new one

With the meta chaotic internal culture, it's hard to handle the cloud as a business. They would be even worse than google cloud